Dynamic Programming Track Before Detect (DP-TBD) is widely applied in the radar detection for weak targets. The algorithm selects the most advantageous state in each stage to integrate the merit function and return the corresponding target status to the merit function that exceeds the detection threshold in the final stage. However, due to the recursive search strategy included in dynamic programming, the constant false alarm rate (CFAR) detection threshold is difficult to be quickly determined by numerical integration. Monte Carlo counting, extreme value theory or its extended algorithm are commonly used to determine the detection threshold of dynamic programming. However, these methods require many times of Monte Carlo simulation, which is time-consuming and difficult to apply to practical projects. In order to achieve fast and accurate acquisition of CFAR detection threshold, this paper proposes an efficient approximation algorithm to determine the DP-TBD CFAR threshold. The algorithm ap-proximates the computational structure of DP-TBD by structural analogy, and then uses the saddle-point approximation method and the numerical integration of the fitting formula to quickly determine the CFAR threshold. Numerical simulation experiments under various conditions and radar field experiments are carried out to verify the high accuracy of the approximate threshold determined by the proposed algorithm. In addition, the algorithm is further extended and can be applied in the noise background with varying power. The corresponding numerical simulation experiments prove the effectiveness of the extended algorithm.

Detection of floating small targets in sea clutter is a challenging problem for maritime surveillance radar. For sea-surface floating targets without regular movement, it is hard for traditional model-based methods to achieve satisfying performance. Considering that the radar echoes will exhibit stronger statistical correlations when the target is present, this work proposes a detector using mean spectral radius (MSR), which is a data-driven method based on the eigenvalue analysis of a random matrix. It is first illustrated that the inner radius of the first-order autoregressive model will decrease with the increase of data correlations when the outer radius is normalized. The spectral radius of the target echo distributes on a smaller value compared with sea clutter due to stronger statistical correlations. The MSR, which is defined to be a specific linear spectral statistics indicating data correlations, is demonstrated to be an effective test statistic to distinguish the target echo from sea clutter. It is shown that by the central limit theorem (CLT), the MSR follows a Gaussian distribution. Experiments on Intelligent PIXel Processing Radar (IPIX) datasets show that the proposed method can effectively improve the detection performance. In addition, the MSR can be applied to construct multiple-feature-based detector to further enhance robustness.

In this paper, we focus on the design of coherent detectors in compound-Gaussian clutter using nonlinear pre-processing system followed by the matched filter (MF). First, to evaluate the effectiveness of pre-processing system, the signal-to-clutter ratio (SCR) gain of pre-processing system is proposed as its performance metric. It is proven that the upper bound of SCR gain is greater than unity for compound-Gaussian clutter, indicating that an appropriate pre-processing system can improve detection performance. Then the suprathreshold stochastic resonance (SSR) system composed of a parallel array of quantizers is introduced, which is an effective and easy-to-implement suboptimal pre-processing system. In compound-Gaussian clutter, a method for determining parameters of quantizer noise, which is purposely injected in the SSR system, is proposed. Finally, the performance improvement of the SSR pre-processing system for weak target in compound-Gaussian clutter is illustrated using both simulated clutter and real sea clutter data. Experiment results show that the proposed method can be applied to detect weak moving targets in sea clutter.

Detectors based on matrix information geometry have been developed recently and demonstrated advantages against conventional methods for detection of targets within nonhomogeneous clutter backgrounds. However, existing methods employing geometric measures to distinguish target and clutter lack sufficient consideration of statistical distribution of covariance matrices on manifold. In this paper, a detector is proposed utilizing the statistical characteristic of covariance matrices on Riemannian manifold, which is modeled by Riemannian Gaussian distribution. A second-order detection problem is formulated utilizing sample covariance matrices. Statistical characteristic of sample covariance matrices on the symmetric positive-definite (SPD) matrix manifold is analyzed to address the detection problem. It is verified that the distribution of sample covariance matrix of multivariate Gaussian observations can be well approximated by Riemannian Gaussian distribution when the number of samples is large. Then the likelihood ratio test (LRT) detector based on Riemannian Gaussian distribution is derived according to the Neyman-Pearson principle. In practical applications, the unknown parameter of clutter distribution in the LRT detector is substituted by the geometric mean of samples in reference cells, which is the maximum likelihood estimate of the mean of Riemannian Gaussian distribution. The above results and the theoretical performance of proposed detector have been validated through numerical simulations. Experiments based on real weather clutter, ground clutter and sea clutter demonstrate that the proposed detector outperform conventional detectors and existing detectors using geometric metrics.

The performance of conventional constant false alarm rate (CFAR) detectors may degrade in nonhomogeneous clutter environments, as accurately estimating the clutter distribution in the cell under test (CUT) using reference cells becomes challenging. In this paper, a CFAR detector based on clutter segmentation with spatial continuity constraints is proposed for target detection within nonhomogeneous weather clutter backgrounds. Analysis of real weather clutter collected by a high-resolution phased array radar indicates that the Rayleigh mixture model can precisely characterize the amplitude distribution of nonhomogeneous weather clutter in spatial domain. The hidden Markov random field (HMRF) model is employed to capture the spatial correlation of weather clutter. Based on this model, clutter segmentation is implemented using the variational expectation-maximization (VEM) algorithm, which provides the posterior class of clutter in each range cell and the estimated parameter of each class. Simulation results indicate that introducing the spatial continuity improves the accuracy of clutter segmentation and parameter estimation. A CFAR detection scheme is proposed, which utilizes the segmentation results to estimate the clutter distribution of the CUT and set the detection threshold accordingly. Experiments conducted using both simulated data and real weather clutter have demonstrated that the proposed method improves detection performance. The proposed method exhibit a maximum increase in detection probability of 8.97% compared to the best-performing benchmark method when the false alarm rate is $10^{-6}$ in real weather clutter.

The recognition of unmanned aerial vehicles (UAVs) has become a core aspect of low-altitude airspace management. Typically, the micro-Doppler characteristics of rotors are employed to accomplish the UAV recognition task. However, the performance of UAV recognition on the basis of micro-Doppler characteristics significantly degrades under low signal-to-noise ratio (SNR) conditions. This paper presents an intelligent small UAV recognition method based on micro-Doppler enhancement to separate UAVs from aerial targets under low SNR conditions. First, a dynamic echo signal model of a UAV rotating blade is established while considering the blade length and rotation frequency parameters. Second, a micro-Doppler characteristic enhancement transformation based on nonlinear body–micro coupling is proposed. In the enhancement algorithm, the phase difference is used to separate the Doppler component of the UAV body and rotor blade. Then, a twodimensional micro-Doppler feature map is created on the basis of the blade length and rotor frequency. Crucially, the microDoppler energy is increased when the values of the parameters match the true values. On the basis of a micro-Doppler feature map, a convolutional neural network (CNN) is constructed to achieve UAV recognition. The effectiveness of the proposed method is verified by experiments using both simulated data and real data. The radar echoes of 2 types of UAVs with S-band carrier frequencies are collected. Among them, DJI-Inspire UAV data with a high SNR of micro-Doppler characteristics are used to verify the correctness of the algorithm, and DJI-Phantom4 UAV data with a low SNR of micro-Doppler characteristics are used to test the performance of the algorithm. The proposed method significantly improves UAV recognition performance under low SNR conditions.